English listening and speaking interaction intelligent training system based on large language model

Abstract

The invention discloses an English listening and speaking interaction intelligent training system based on a large language model, and relates to the technical field of English learning, and the training system comprises a user capability graph module, an interaction feedback module, a personalized learning path generation module, a scene simulation module, an interaction rhythm control module and a community cooperative training module. According to the method, the pragmatic ability, the cross-language communication ability and the linguistic ability are jointly incorporated into the training content, and the adaptive training content is generated in combination with a real cross-language communication scene, so that a user synchronously masters expression specifications and culture adaptation key points in different scenes in the training process; the effect that training content and real communication requirements are deeply matched is achieved, the user is helped to accurately adapt to a native English scene and real phrase habits of people, the requirements of an actual application scene are met, and substantial improvement from basic language knowledge mastering to real communication ability is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of English learning, in particular to an English listening and speaking interactive intelligent training system based on a large language model. BACKGROUND

[0002] English, as an international official language, is used in many occasions, and the number of people learning English in non-English-speaking countries is increasing. It is crucial to evaluate the listening and speaking abilities in English learning, which directly reflects the level of English. English listening and speaking interactive training is a comprehensive training method for improving English listening and speaking abilities through scenario simulation, role-playing, cooperative activities, etc. The core is to promote the coordinated development of language input and output through simulated communication scenarios.

[0003] Currently, the English listening and speaking training system only focuses on the evaluation of language ability (pronunciation, grammar and vocabulary), and does not include pragmatic ability and cross-language communication ability in the quantitative system of training, resulting in a gap between training content and real cross-language communication scenarios. Users can master basic language knowledge, but it is difficult to accurately adapt to the real language habits of English-speaking environment groups, resulting in the problem of being able to speak but not being able to speak correctly. It cannot meet the needs of actual application scenarios, resulting in a mismatch between training effectiveness and real communication ability improvement, and it is difficult to achieve effective breakthroughs in comprehensive listening and speaking application ability. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides an English listening and speaking interactive intelligent training system based on a large language model, which solves the problems mentioned in the background art.

[0005] To achieve the above purpose, the present application realizes the following technical scheme: an English listening and speaking interactive intelligent training system based on a large language model, characterized in that: the training system comprises: a user ability graph module for collecting voice data and text data of the user in the process of listening and speaking English interaction, quantitatively evaluating the language ability, pragmatic ability and cross-language communication ability of the user through the built-in language large model, generating a user ability graph, and the evaluation weight of the user ability graph can be adaptively adjusted according to the user's learning goal; an interactive feedback module for analyzing and identifying the error content of the user's English listening and speaking according to the voice data and text data of the user in the process of listening and speaking English interaction, and outputting interactive feedback data including pronunciation correction, pragmatic optimization suggestion, cross-language language adaptation prompt and dictation accuracy; a personalized learning path generation module configured to automatically generate a long-term personalized learning scheme based on historical training data of user historical English listening and speaking, the user ability graph, and the interaction feedback data, the personalized learning scheme supporting user-defined adjustment parameters; a scenario simulation module configured to generate an interactive dialogue task in accordance with pragmatic norms by combining scenario elements based on the user ability graph module and a preset cross-language scenario library, and to check the pragmatic deviation of cross-language in user expression; an interaction rhythm control module configured to adjust the difficulty of the interactive dialogue according to the user ability graph, and to adjust the interaction rhythm by extracting emotion features through voice signal processing and using a sentiment recognition mechanism to determine the user's emotional state; a community collaborative training module configured to match a corresponding user community according to the user ability graph, the interaction feedback data, and the personalized learning scheme of the user, the user community regularly holds English listening and speaking simulation scenarios, the simulation scenarios are hosted, judged, and logically corrected by the language large model, and a corresponding training report is generated after the simulation scenario ends.

[0006] Preferably, the quantitative evaluation indicators of the user ability graph module specifically include: the language ability includes pronunciation accuracy, grammar correctness, vocabulary richness, and logical coherence; the pragmatic ability includes scene adaptation, expression appropriateness, and topic continuity; the cross-language communicative competence includes cross-language context adaptation, cultural symbol recognition rate, and inappropriate language avoidance rate; the index weights of the language ability, the pragmatic ability, and the cross-language communicative competence are allocated in the range of 0.1-0.4 according to user learning goals.

[0007] Preferably, the output content of the interaction feedback module is specifically: the pronunciation correction includes feature comparison data between user voice and standard voice and pronunciation part guidance instructions, the pragmatic optimization suggestion provides 2-3 scenario-based alternative expressions, the cross-language language adaptation prompt marks the key points of the target language cultural background association, and the dictation accuracy is presented in percentage form, with highlighted words or sentence patterns of dictation errors.

[0008] Preferably, the reinforcement learning model of the personalized learning path generation module is specifically designed as: the state space includes user three-dimensional ability scores, knowledge point mastery rates, and learning goal completion progress; the action space includes scene difficulty adjustment, training duration allocation, and knowledge point reinforcement priority setting; The reward value is calculated as follows: Reward value = 0.3 × Ability improvement rate + 0.4 × Goal achievement rate + 0.3 × Task completion efficiency.

[0009] Preferably, the cross-language scene library of the scenario simulation module includes 8 core scene categories, specifically: Scenarios including academic exchanges, business communication, daily social interactions, travel, workplace collaboration, exam preparation, cross-cultural interviews, and public speaking; The elements of the scenario include: dialogue roles, dialogue background, pragmatic rules, and usage habits. The pragmatic deviation verification is performed by matching and verifying user expressions with the pragmatic norms of the scenario through the language big model.

[0010] Preferably, the emotional features extracted by the interactive rhythm control module include: speech rate, tone, short-term volume and zero-crossing rate. The emotion recognition mechanism identifies four types of emotional states of the user: tension, confusion, confidence and fatigue. The corresponding adjustment strategies are as follows: when nervous, extend the thinking time by 5-10 seconds; when confused, add guiding statements; when confident, increase the difficulty level of the dialogue; and when tired, reduce the duration of a single training session and remind the participant to rest.

[0011] Preferably, the matching dimensions of the community collaborative training module include: the similarity of the user ability graph, the fit of learning objectives, the complementarity of weaknesses in the interactive feedback data, and the overlap of training periods. The size of the community is 3-10 people, and the simulation duration of the community is 30-60 minutes.

[0012] Preferably, the user capability graph module is associated with the personalized learning path generation module. When the score of the corresponding capability dimension in the user capability graph increases by ≥10%, the personalized learning path generation module adjusts the training weight of the corresponding capability dimension, reduces the proportion of repetitive training, and increases the proportion of advanced tasks.

[0013] Preferably, the interactive feedback data output of the interactive feedback module includes: The system offers text descriptions, voice demonstrations, and visual data charts. Users can choose the feedback presentation format and frequency, which can be real-time feedback, single-round end-of-round feedback, or daily summary feedback.

[0014] Preferably, the cycle of the personalized learning program is 1-3 months. The personalized learning program includes: scenario training sequence, knowledge point reinforcement, ability improvement goals and phased assessments. Users can customize and modify the target node time, adjust the priority of scenario training and add or remove special training tasks.

[0015] This invention provides an intelligent training system for interactive English listening and speaking based on a large language model. It has the following beneficial effects: (1) By incorporating pragmatic ability and cross-language communication ability into the training content together with language ability, combining with real cross-language communication scenarios to generate adaptive training content, allowing users to master expression norms and cultural adaptation points in different scenarios during the training process, achieving the effect of deep integration of training content and real communication needs, helping users accurately adapt to the real language habits of English native speaker scenarios and populations, meeting the needs of actual application scenarios, and realizing substantial improvement from basic language knowledge mastery to real communication ability.

[0016] (2) Through scenario simulation and targeted interaction feedback, focusing on the specialized training of pragmatic adaptation and cross-language communication ability, guiding users to understand the language logic and cultural habits of English native speaker environment in interaction, making the training process and real communication situation coincide with each other, solving the limitation of traditional training only limited to basic language knowledge, allowing users not only to master core elements such as pronunciation and grammar, but also to form scenario-based expression thinking, improving the appropriateness and accuracy of expression, making the training effect consistent with the demand for real communication ability improvement, and effectively promoting the overall advancement of comprehensive listening and speaking application ability.

[0017] (3) Abandoning the single focus on language ability training mode, through coordination to realize the coordinated cultivation of language, pragmatic and cross-language communication ability, and setting up real social group simulation training, allowing users to simultaneously strengthen the ability of basic language use, scenario adaptation expression, and cross-cultural communication adaptation in training, achieving the effect of comprehensive ability improvement in all directions, changing the status quo of traditional training and real cross-language communication scenarios, effectively bridging the gap between training and actual application, helping users get rid of the problem of only being able to express basic expressions and being difficult to apply flexibly, realizing effective breakthrough of comprehensive listening and speaking application ability, and meeting the communication needs in different actual scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0018] Fig. 1 The learning flowchart of the English listening and speaking interactive intelligent training system based on a large language model of the present application; Fig. 2 The training process flowchart of the English listening and speaking interactive intelligent training system based on a large language model of the present application; Fig. 3 The social group training flowchart of the English listening and speaking interactive intelligent training system based on a large language model of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present application.

[0020] Embodiment 1 See Figs. 1-3 The present application provides an English listening and speaking interactive intelligent training system based on a large language model. To achieve the above purpose, the present application is implemented through the following technical solutions: the training system comprises: A user ability graph module is configured to collect voice data and text data of the user in the process of English listening and speaking interaction, quantitatively evaluate the language ability, pragmatic ability and cross-language communication ability of the user through a built-in large language model, and generate a user ability graph. The evaluation weight of the user ability graph can be adaptively adjusted according to the user's learning goal. An interaction feedback module is configured to analyze and identify the error content in the user's English listening and speaking according to the voice data and text data of the user in the process of English listening and speaking interaction, and output interaction feedback data including pronunciation correction, pragmatic optimization suggestion, cross-language language adaptation prompt and dictation accuracy. A personalized learning path generation module is configured to generate a long-term personalized learning scheme automatically by combining a reinforcement learning model to construct a reward mechanism according to the historical training data of the user's historical English listening and speaking, the user ability graph and the interaction feedback data. The personalized learning scheme supports user-defined adjustment parameters. A scenario simulation module is configured to combine scenario elements to generate an interactive dialogue task that conforms to the pragmatic specification based on the user ability graph module and a preset cross-language scenario library, and to check the pragmatic deviation of the user's expression. An interaction rhythm control module is configured to adjust the difficulty of the interactive dialogue according to the user ability graph, extract emotion features through voice signal processing, and adjust the interaction rhythm by using a sentiment recognition mechanism to judge the user's emotional state. A social collaborative training module is configured to match a corresponding user community according to the user ability graph, the interaction feedback data and the personalized learning scheme of the user, and regularly open English listening and speaking simulation scenarios in the user community. The simulation scenarios are hosted, judged and logically corrected by the language model, and a corresponding training report is generated after the simulation scenario ends.

[0021] In this embodiment, after the user logs in the system, the user sets the goal to be achieved by training and learning, the system initializes the parameters of each module, completes the loading of core resources such as cross-language scene library, pragmatic rule library and language large model, and triggers the initial interaction task to the user (such as a brief self-introduction and a scene-based topic response) by the user ability graph module. Real-time collection of voice data and text data in the user interaction process (voice data is converted into text in real time, and the system broadcasts the dialogue is transcribed into text by the user), the built-in language large model analyzes the collected data in multiple dimensions, respectively quantifies and evaluates the language ability, pragmatic ability and cross-language communication ability, and generates an initial user ability graph by combining the user-set learning goal with the adaptive allocation of the evaluation weight of the three types of abilities to establish the user ability baseline; The personalized learning path generation module calls the user initial ability graph, the system preset language learning rule model, and simultaneously associates the user's potential learning preference (based on the learning goal deduction), constructs a dynamic reward mechanism through a reinforcement learning model; integrates the above data and mechanism, automatically generates a long-term personalized learning scheme containing scene training sequence, knowledge point strengthening direction, training cycle planning and ability improvement target, and the user can customize and adjust the parameters in the scheme (such as training frequency, scene priority, etc.) according to the user's own needs.

[0022] The scene simulation module takes the user initial ability graph as the core basis, selects the appropriate scene elements (dialogue roles, scene backgrounds, pragmatic rules, and language habits) from the cross-language scene library, constructs an interactive dialogue task with coherence and authenticity through a scene logic combination algorithm, ensures that the task difficulty and pragmatic specification are highly consistent with the user's ability level and learning goal, pushes the constructed scene-based interactive task to the user, and integrates it into the personalized learning scheme. Subsequently, the listening and speaking interactive training can be started.

[0023] During the training process, the interaction rhythm control module synchronously collects the user's voice data stream, extracts emotion features such as speech rate, tone, and short-term volume through voice signal processing technology, uses emotion recognition mechanism to judge the user's emotional state (nervous, confused, confident, tired, etc.) in real time, and dynamically adjusts the difficulty of the interactive dialogue (such as vocabulary complexity, sentence structure, and topic depth) by referring to the core score in the user ability graph. If negative emotions (nervous, confused, and tired) are detected, corresponding adjustment strategies are triggered, such as extending the thinking time, supplementing the guiding sentences, and shortening the single-round training duration. If the user's emotional state is good and the ability is suitable, the interaction rhythm and task challenge can be moderately improved to ensure the adaptability and fluency of the training process.

[0024] During the training process, the interactive feedback module receives voice data and text data in real time during user interaction, analyzes the acoustic features and prosodic features of the voice data, analyzes the semantic structure, expression logic, and pragmatic adaptability of the text data, constructs a multi-level error recognition model, accurately locates pronunciation deviations, pragmatic errors, cross-language adaptation errors, and other problems, and generates comprehensive interactive feedback data containing pronunciation correction (feature comparison and pronunciation guidance), pragmatic optimization suggestions (contextualized alternative expressions), cross-language expression adaptation prompts (cultural background association), and transcription accuracy annotations (error highlighting) for the identified error content. The user ability graph module receives error data output by the interactive feedback module, real-time performance data during the interaction process, and combines the initial ability graph baseline to quantitatively evaluate the user's language ability, pragmatic ability, and cross-language communication ability through a language large model, dynamically updates the scores of various indicators in the user ability graph, and reflects the user's ability changes in real time to provide accurate data support for subsequent training adjustments.

[0025] The personalized learning path generation module integrates the updated user ability graph, historical training data, and interactive feedback data, optimizes the core parameters of the reward mechanism through a reinforcement learning model, dynamically adjusts the personalized learning scheme (such as increasing the proportion of scene training corresponding to weak items, increasing the proportion of advanced tasks for already-meeting-standards abilities, and optimizing training cycle nodes) according to the user's ability improvement, error improvement effect, and training task completion, and forms a closed-loop training logic of "training-evaluation-feedback-optimization"; After completing the daily training content and the customized training scheme, the social group collaborative training module calls the user's real-time ability graph, interactive feedback data, and personalized learning scheme, selects suitable users through multi-dimensional data comparison algorithms (ability similarity, target matching degree, weak item complementarity, and training time period coincidence), forms a training social group with reasonable size, and ensures that the members of the social group have synergy and complementarity in training needs and ability levels; The system regularly pushes scenario simulation tasks to the formed social group, and the language large model acts as the scenario host to guide the listening and speaking interaction training of the social group users; during the training process, the language large model analyzes the expression content of each user in real time, makes listening and speaking judgments (pronunciation accuracy and expression logic), and corrects language logic, timely resolves pragmatic conflicts and topic deviations between users, and ensures the orderly progress of the social group training; After the social group scenario training is completed, the language large model integrates the interactive data, ability performance, error distribution, and improvement space of each user to generate an individualized social group training report (including personal performance analysis, group synergy review, and improvement suggestions); at the same time, the effective data in the social group training is supplemented to the cross-language scenario library and the pragmatic rule library, realizing the iterative optimization of system resources; The above process is repeated, and the user ability graph and personalized learning scheme are updated synchronously after each round of scene training (individual / group) or a training cycle. The data of individual training and group training are continuously integrated to optimize the model and expand the resource library, realize the synchronous iteration of user comprehensive listening and speaking ability and system service ability, and promote the user to gradually achieve the preset learning goal.

[0026] Embodiment 2 Specifically, the quantitative evaluation indicators of the user ability graph module include: The language ability includes: pronunciation accuracy, grammar correctness, vocabulary richness, and logical coherence; The pragmatic competence includes: scene adaptation, expression appropriateness, and topic continuity; The cross-language communication ability includes: cross-language context adaptation, cultural symbol recognition rate, and inappropriate language avoidance rate; The index weight of language ability, pragmatic competence and cross-language communication ability is allocated in the range of 0.1-0.4 according to the user learning goal.

[0027] The output content of the interactive feedback module is specifically: The pronunciation correction includes: feature comparison data of user voice and standard voice and pronunciation part guidance instructions, pragmatic optimization suggestions provide 2-3 sceneized alternative expressions, cross-language language adaptation prompt marks the key points of the target language cultural background association, and the listening accuracy is presented in percentage form, while highlighting the wrong words or sentence patterns in the listening.

[0028] The reinforcement learning model of the personalized learning path generation module is specifically designed as: The state space includes: user three-dimensional ability score, knowledge point mastery rate, and learning goal completion progress; The action space includes: scene difficulty adjustment, training duration allocation, and knowledge point reinforcement priority setting; The reward value calculation method of the reward mechanism is: reward value=0.3×ability improvement amplitude+0.4×target achievement rate+0.3×task completion efficiency; In this embodiment, the evaluation of pronunciation accuracy is based on the deep comparison of voice acoustic features and standard voice model, combined with the semantic association analysis of language large model on pronunciation details, the grammar correctness is realized through the grammar rule library and real-time semantic analysis ability of language large model, the judgment of sentence structure and grammar application, the vocabulary richness focuses on the diversity, adaptability and semantic accuracy of the vocabulary in user expression, and the logical coherence is based on the comprehensive evaluation of text semantic association degree, sentence connection rationality and expression logical integrity; The pragmatic competence indicators revolve around the fit between user expression and scenario needs and role identity. They are quantified through scenario pragmatic rule matching of the language big model. The cross-language communicative competence indicators focus on the user's mastery and application of the target language's cultural background, symbol meaning, and expression taboos. The weight allocation is determined by the model's semantic analysis of the user's learning objectives, which determines the priority of the core competence dimensions and optimizes them dynamically. The pronunciation correction feature comparison covers a multi-dimensional analysis of acoustic and prosodic features. The pronunciation guidance is based on the adaptation of the movement principle of the articulatory organs in linguistics to the target pronunciation. The pragmatic optimization suggestions provide contextualized alternative expressions, which are generated by a large language model based on the pragmatic norms of the current interaction scenario and the core semantics of the user's expression. This ensures that the alternatives meet the requirements of the scenario while preserving the user's original meaning. The cross-language usage adaptation prompts use a knowledge base that associates with the target language's cultural background to mark cultural customs, communication rules, and potential semantic differences related to the current expression, helping users understand the cultural logic behind the expression. The presentation of dictation accuracy and error marking accurately locates the error position based on text comparison technology and highlights the content that needs to be strengthened. The parameters in the state space are dynamically updated based on the user's historical training data and real-time evaluation results, comprehensively and realistically reflecting the user's learning status and ability level. The adjustment strategies in the action space are based on the changes in the parameters of the state space. The reinforcement learning model determines the optimal adjustment direction through continuous iterative learning, ensuring the adaptability of the training plan to the user's learning. The core design of the reward mechanism is to balance the quality of ability improvement, the efficiency of goal advancement, and the learning experience. By allocating weights, the core training requirements are highlighted, guiding the model to generate a learning path that better meets the user's improvement needs, while also motivating the user to maintain the continuity and enthusiasm of training.

[0029] Example 3 Specifically: The cross-language scenario library of the scenario simulation module includes 8 core scenarios, namely: Scenarios include academic exchanges, business communication, daily social interactions, travel, workplace collaboration, exam preparation, cross-cultural interviews, and public speaking. The elements of a scenario include: dialogue roles, dialogue background, pragmatic rules, and usage habits. Pragmatic deviation verification uses a large language model to match and verify user expressions with the pragmatic norms of the scenario.

[0030] The emotional features extracted by the interactive rhythm control module include speech rate, tone, short-term volume and zero-crossing rate. The emotion recognition mechanism identifies four types of emotional states of the user: tension, confusion, confidence and fatigue. The corresponding adjustment strategies are as follows: when nervous, extend the thinking time by 5-10 seconds; when confused, add guiding statements; when confident, increase the difficulty level of the dialogue; and when tired, reduce the duration of a single training session and remind the participant to rest.

[0031] The interaction feedback data output of the interaction feedback module comprises: Text description, voice demonstration and visual data chart, the user can independently select the feedback presentation form and the feedback frequency, the feedback frequency is real-time feedback, single round end feedback and daily summary feedback; In this embodiment, the construction of the cross-language scenario library is based on the investigation and analysis of real cross-language communication scenarios, covering the types of scenarios frequently used by users, and the design of scenario elements strictly follows the pragmatic rules, cultural customs and communication habits of the target language, ensuring the authenticity and practicality of the scenarios, and the combination of scenario elements is dynamically adapted according to the user ability map and learning objectives, generating interactive dialogue tasks with logical coherence and training pertinence. In the pragmatic bias checking process, the language large model first deeply analyzes the pragmatic specifications, role positioning and core needs of the current scenario, and then performs semantic and pragmatic analysis on the user's expression, accurately identifying the expression content and deep reasons that do not match the scene specifications; The extraction of emotional features is realized by multi-dimensional analysis of the original voice data through voice signal processing, and the core feature parameters that can truly reflect the user's emotional state are obtained. The emotional recognition mechanism is based on the comparison of pre-set emotional feature model and real-time features, which realizes the accurate judgment and dynamic tracking of user's emotional state. The association logic between emotional state and adjustment strategy is based on the dual consideration of user learning experience and training efficiency, which dynamically adjusts the interaction rhythm, effectively alleviates the negative emotions of users, reduces the training pressure, and ensures the smooth progress of the training process and the efficient achievement of the training target; The text description of the interaction feedback data output focuses on clear analysis of the problem and operable optimization suggestions, the user can intuitively view the feedback suggestions, provide standard and normative expression reference for voice demonstration, and facilitate the user to intuitively perceive the rhythm and tone of correct expression; The visual data chart presents the user's ability performance, error distribution and improvement trend in an intuitive form, helping the user to quickly grasp the training effect. The autonomous selection of feedback presentation form meets the learning habits and reception preferences of different users, and the diversified design of feedback frequency adapts to the needs of different training scenarios. Real-time feedback is suitable for training links that need immediate correction, single round end feedback is convenient for users to review after single round training, and daily summary feedback helps users to comprehensively understand the training effect and overall improvement direction of the day.

[0032] Embodiment 4 Specifically, the matching dimensions of the community collaborative training module include: the similarity of the user ability map, the learning target matching degree, the complementary degree of weak items in the interaction feedback data, and the training time period coincidence degree. The size of the community is 3-10 people, and the simulation time of the community is 30-60 minutes.

[0033] The user ability map module is associated with the personalized learning path generation module, when the score improvement range of the corresponding ability dimension in the user ability map is greater than or equal to 10%, the personalized learning path generation module adjusts the training weight of the corresponding ability dimension, reduces the proportion of repeated training, and increases the proportion of advanced tasks.

[0034] The single cycle of the personalized learning scheme is 1-3 months, and the content of the personalized learning scheme includes: scene training sequence, knowledge point reinforcement, ability improvement target and stage evaluation, and the user can customize and modify the target node time, adjust the priority of scene training and increase or decrease special training tasks; In the embodiment, the parameters of the matching dimension are quantitatively analyzed by a multi-source data comparison algorithm, the similarity of the user ability map ensures that the ability level of the community members is in the adaptive interval, avoiding that the large ability gap affects the training effect, the learning target fit degree ensures the consistency of the community training direction, ensuring that the members carry out collaborative training under the same core demand, the weak item complementary degree is promoted by analyzing the short board distribution in the interaction feedback data, promoting the mutual learning and complementary advantages of the community members in training, and the training period coincidence degree ensures that the community training can be carried out smoothly, reducing time conflicts; The setting of the community size is based on the double consideration of interactive participation and training efficiency, which ensures that each member can fully participate in interaction, obtain enough expression and feedback opportunities, and ensures the orderly progress of the training process, the simulation time length design takes into account the training depth and user energy, ensuring that the training target is achieved within the effective time, avoiding that too long or too short affects the training effect; The matching mechanism of the community collaborative training module accurately adapts to the ability level, learning needs and time arrangement of the user, the linkage between modules ensures that the learning path can dynamically follow the user's ability improvement pace, the personalized learning scheme has scientificity, flexibility and autonomy, enabling users to realize complementary advantages and common progress in community collaboration, efficiently break through the ability short board in the dynamically optimized learning path, enhance the autonomy and adaptability of training through the self-defined adjustment function, effectively improve the comprehensive effect and user experience of training, and accelerate the comprehensive and stable improvement of comprehensive listening and speaking ability.

[0035] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application.

Claims

1. An intelligent training system for interactive English listening and speaking based on a large language model, characterized by: The training system includes: The user competence graph module is used to collect voice and text data of users during English listening and speaking interactions. It uses a built-in language model to quantitatively evaluate users’ language competence, pragmatic competence and cross-language communicative competence, and generate a user competence graph. The evaluation weights of the user competence graph can be adaptively adjusted according to the user’s learning goals. The interactive feedback module is used to analyze and identify errors in the user's English listening and speaking based on the voice data and text data during the user's English listening and speaking interaction, and output interactive feedback data including: pronunciation correction, pragmatic optimization suggestions, cross-language terminology adaptation prompts, and dictation accuracy. The personalized learning path generation module is used to automatically generate a long-term personalized learning plan based on the user's historical English listening and speaking training data, the user's ability graph, and the interaction feedback data, combined with a reinforcement learning model to build a reward mechanism. The personalized learning plan supports user-defined parameter adjustments. The scenario simulation module, based on the user capability graph module and the preset cross-language scenario library, is used to combine scenario elements to generate interactive dialogue tasks that conform to pragmatic norms and to verify cross-language pragmatic deviations in user expressions. The interaction rhythm control module is used to adjust the difficulty of the interactive dialogue according to the user's ability map, and at the same time extract emotional features through voice signal processing, use the emotion recognition mechanism to judge the user's emotional state, and adjust the interaction rhythm accordingly. The community collaborative training module is used to match corresponding user communities based on the user's user ability graph, the interaction feedback data, and the personalized learning plan. The user community regularly hosts English listening and speaking simulation scenarios. The simulation scenarios are hosted, listening and speaking judgments are made, and language logic is corrected by the language big model. A corresponding training report is generated after the simulation scenario ends.

2. The English listening and speaking interactive intelligent training system based on a large language model according to claim 1, characterized in that: The quantitative evaluation indicators of the user capability graph module specifically include: The language abilities mentioned include: pronunciation accuracy, grammatical correctness, vocabulary richness, and logical coherence. The pragmatic capabilities mentioned include: context suitability, appropriateness of expression, and topic continuity; The cross-language communicative competence includes: cross-language context adaptability, cultural symbol recognition rate, and inappropriate language avoidance rate; The weights of the indicators of language competence, pragmatic competence, and interlingual communicative competence are allocated in the range of 0.1-0.4 according to the user's learning objectives.

3. The English listening and speaking interactive intelligent training system based on a large language model according to claim 1, characterized in that: The specific output of the interactive feedback module is as follows: The pronunciation correction includes: feature comparison data of user speech and standard speech and guidance on pronunciation positions; the pragmatic optimization suggestions provide 2 to 3 scenario-based alternative expressions; the cross-language terminology adaptation prompts highlight the key points related to the target language cultural background; the dictation accuracy is presented as a percentage, while highlighting words or sentence structures that are dictated incorrectly.

4. The English listening and speaking interactive intelligent training system based on a large language model according to claim 1, characterized in that: The reinforcement learning model of the personalized learning path generation module is specifically designed as follows: The state space includes: user's three-dimensional ability score, knowledge point mastery rate, and progress in achieving learning objectives; The action space includes: scene difficulty adjustment, training time allocation, and knowledge point reinforcement priority settings; The reward value is calculated as follows: Reward value = 0.3 × Ability improvement rate + 0.4 × Goal achievement rate + 0.3 × Task completion efficiency.

5. The English listening and speaking interactive intelligent training system based on a large language model according to claim 1, characterized in that: The cross-language scenario library of the scenario simulation module includes eight core scenarios, specifically: Scenarios including academic exchanges, business communication, daily social interactions, travel, workplace collaboration, exam preparation, cross-cultural interviews, and public speaking; The elements of the scenario include: dialogue roles, dialogue background, pragmatic rules, and usage habits. The pragmatic deviation verification is performed by matching and verifying user expressions with the pragmatic norms of the scenario through the language big model.

6. The English listening and speaking interactive intelligent training system based on a large language model according to claim 1, characterized in that: The emotional features extracted by the interactive rhythm control module include: speech rate, tone, short-term volume and zero-crossing rate. The emotion recognition mechanism identifies four types of emotional states of the user: tension, confusion, confidence and fatigue. The corresponding adjustment strategies are as follows: when nervous, extend the thinking time by 5-10 seconds; when confused, add guiding statements; when confident, increase the difficulty level of the dialogue; and when tired, reduce the duration of a single training session and remind the participant to rest.

7. The English listening and speaking interactive intelligent training system based on a large language model according to claim 1, characterized in that: The matching dimensions of the community collaborative training module include: the similarity of the user ability graph, the fit of learning objectives, the complementarity of weaknesses in the interaction feedback data, and the overlap of training periods. The community size is 3-10 people, and the simulation duration of the community is 30-60 minutes.

8. The English listening and speaking interactive intelligent training system based on a large language model according to claim 1, characterized in that: The user capability graph module is associated with the personalized learning path generation module. When the score of the corresponding capability dimension in the user capability graph increases by ≥10%, the personalized learning path generation module adjusts the training weight of the corresponding capability dimension, reduces the proportion of repetitive training, and increases the proportion of advanced tasks.

9. The English listening and speaking interactive intelligent training system based on a large language model according to claim 1, characterized in that: The interactive feedback data output of the interactive feedback module includes: The system offers text descriptions, voice demonstrations, and visual data charts. Users can choose the feedback presentation format and frequency, which can be real-time feedback, single-round end-of-round feedback, or daily summary feedback.

10. The English listening and speaking interactive intelligent training system based on a large language model according to claim 1, characterized in that: The personalized learning program has a single-cycle period of 1-3 months. The personalized learning program includes: scenario training sequences, knowledge point reinforcement, ability improvement goals and phased assessments. Users can customize and modify the target node time, adjust the priority of scenario training, and add or remove special training tasks.

Images